Advertisement
source and destination IP address as well as its source and destination TCP/UDP ports. To explain this further, if a user were to start up two
applications—for example, an e-mail client and an FTP session—each session would kick off a set of unique flows. In this mode, if a microflow
policer were applied with a limit set to 2 Mb, then the e-mail session would be limited to 2 Mb, and the FTP session would be limited to 2 Mb.
This would equate to a total of 4 Mb of traffic. Comparing this to an aggregate policer of 2 Mb, then the combined volume of traffic from the FTP
and e-mail sessions would be limited to 2 Mb.
Policing: User-Based Rate Limiting
User-based rate limiting (UBRL) is an enhancement to microflow policing introduced with the PFC3. It provides a configuration option to change
the way in which a flow is viewed by the system. In the previous section, the example showed that for the user initiating the FTP and e-mail
applications, two discrete flows would be seen by the system. In this sense, each flow would be limited to the stated rate. UBRL takes advantage of
a new enhancement in the PFC that allows a flow to be viewed as everything originating from a unique source or destination IP address. In technical
speak, this enhancement is known as a source IP only flow mask or destination IP only flow mask. What this means is that a microflow policer can
now be applied to limit traffic originating to or from each user. It allows the administrator to put in place some rules that allow policies limiting
traffic on a per-user basis, something microflow policing was not able to do on earlier PFC models. Using the preceding example, if each user
initiated multiple sessions (e-mail, Telnet, FTP, HTTP, and so on), each user (all data for that user) would be limited to 2 Mb of data.
Congestion Avoidance: Tail Drop
As a switch port queue begins to fill with data, thresholds can be used to identify what traffic can be dropped when the threshold is breached.
A packet is primed with a priority value, and the priority value identifies with which threshold this packet is going to be associated. When that
threshold is breached, any packet arriving at the queue with that priority value will be dropped. Packets with that priority value will continue to
be dropped while the amount of data in the queue exceeds that threshold. Figure 9 provides a pictorial view of how thresholds are viewed on a
given queue.
Figure 9. Tail Drop
Congestion Avoidance: WRED
WRED is less aggressive than tail drop, and it targets fewer flows when it initially begins its drop process. When the first (low) threshold is
exceeded, the WRED algorithm will start to randomly drop packets tagged with a particular priority value. The algorithm will attempt to minimize
the impact to multiple flows by only targeting a few select flows. As the queue continues to approach the second threshold the WRED algorithm
begins to more aggressively drop data, and more flows are susceptible to having packets dropped. (See Figure 10.)
© 2005 Cisco Systems, Inc. All rights reserved.
Important notices, privacy statements, and trademarks of Cisco Systems, Inc. can be found on cisco.com.
Page 9 of 18
Advertisement
